New AI model predicts sudden cardiac death risk better in cardiac sarcoidosis
Study shows tool outperformed guideline measures to identify high-risk patients
Written by |
A new artificial intelligence (AI) model predicted sudden cardiac death (SCD) risk more accurately than current clinical guidelines in people with cardiac sarcoidosis, according to a U.S. study.
The AI model, called Multimodal Artificial Intelligence for Ventricular Arrhythmia Risk Stratification in Cardiac Sarcoidosis (MAARS-CS), estimates risk using heart MRI scans and key clinical information.
“The model’s robustness and interpretability enhance its potential as a reliable clinical decision support tool,” the researchers wrote. “With further validation, MAARS-CS may improve personalized patient care in sarcoidosis.”
The study, “Predicting Sudden Cardiac Death in Patients With Sarcoidosis Using a Multimodal Artificial Intelligence Model,” was published in JACC: Clinical Electrophysiology.
What cardiac sarcoidosis is and why it raises sudden death risk
Sarcoidosis is caused by the formation of granulomas, or small clusters of inflammatory cells, in different tissues and organs. When the condition affects the heart, known as cardiac sarcoidosis, it can disrupt normal heart rhythms and raise the risk of sudden cardiac death.
For people at high risk, doctors may recommend an implantable cardioverter-defibrillator (ICD) — a device that delivers electrical pulses to stop dangerous heart rhythms — to help reduce the risk of death.
However, current guidelines for deciding who should receive an ICD are imperfect. As the researchers noted, they can lead to some lower-risk patients receiving ICDs without benefit, while some higher-risk patients remain unprotected. “Therefore, developing accurate SCD risk stratification approaches in patients with CS [cardiac sarcoidosis] is of paramount clinical importance,” they wrote.
To address this gap, the researchers developed an AI model to estimate SCD risk in people with cardiac sarcoidosis. The MAARS-CS model was developed using cardiovascular magnetic resonance with late gadolinium enhancement (CMR-LGE) images, along with clinical data. CMR-LGE is a specialized heart MRI technique that uses a contrast dye to highlight areas of scarring or damage in the heart muscle.
The goal was to identify patterns linked to dangerous heart rhythms and sudden death risk, helping clinicians better assess who may need preventive treatment.
The model was tested in 317 people with probable or definite cardiac sarcoidosis evaluated at Johns Hopkins Hospital. Participants had a mean age of 53.3 years, 56% were women, and they were followed for about 8.5 years.
How often sudden cardiac events occurred
Overall, 39 patients (12%) experienced serious heart rhythm–related events linked to sudden cardiac death. Composite events included sustained ventricular tachycardia, ventricular fibrillation (a life-threatening condition where ventricles quiver rather than pump), appropriate defibrillator shock (an electrical jolt to the heart to correct abnormal heart rhythms), antitachycardia pacing, and SCD.
Compared with those who did not experience these events, patients who did were more often men (64% vs. 41%) and more likely to have a history of heart failure (51% vs. 17%). They also had a significantly lower median left ventricular ejection fraction (LVEF) — a measure of how well the heart pumps blood — 52% compared with 62% in those without events.
The researchers first evaluated how well LVEF alone predicted SCD risk in this patient group. Using the standard guideline of 35% or lower, LVEF separated higher- and lower-risk patients with a balanced accuracy of 59%. Raising the threshold to 50% or 57% improved balanced accuracy to 68% and 73%, respectively.
“These findings suggest that patients with CS with mild to moderate impairment of LVEF are also at elevated risk for SCD,” the researchers wrote.
Compared with LVEF alone, MAARS-CS showed clearer separation between patients who did and did not experience events, indicating that “MAARS-CS stratified risk among patients with CS more effectively than LVEF,” the researchers added.
Overall, MAARS-CS achieved a mean balanced accuracy of 83%, outperforming continuous LVEF, which reached 77%.
How the AI model could change patient care
In a clinical context, MAARS-CS identified 26 additional high-risk patients who would have been missed using the standard LVEF cutoff of 35% or lower. As a result, the AI model reduced the proportion of high-risk patients missed by LVEF alone by 67%.
“Moreover, MAARS-CS maintained robust risk prediction accuracy across different image qualities and different sequences of LGE-CMR scans,” the team wrote. The model “provides interpretability by highlighting key risk factors and image regions influencing its predictions, underscoring the model’s potential as a reliable clinical decision support tool.”
The researchers cautioned that “further studies are warranted to validate the robustness of this model in diverse clinical settings and access how its incorporation into clinical practice may impact patient outcomes.”
